dolfin team mailing list archive

[Johan Hake <hake@xxxxxxxxx>]

On Monday 13 October 2008 18:34:01 Ostien, Jakob T wrote:
> > -----Original Message-----
> > From: dolfin-dev-bounces@xxxxxxxxxx
> > [mailto:dolfin-dev-bounces@xxxxxxxxxx] On Behalf Of Anders Logg
> > Sent: Monday, October 13, 2008 8:37 AM
> > To: dolfin-dev@xxxxxxxxxx
> > Subject: Re: [DOLFIN-dev] BC specification in PyDOLFIN
> >
> > On Mon, Oct 13, 2008 at 08:10:35AM -0600, Ostien, Jakob T wrote:
> > > Hi,
> > >
> > > Is it currently possible to set the boundary condition
> >
> > search specification in PyDOLFIN?
> >
> > > in cpp
> > > DirichletBC bc(function, mesh, subdomain, geometric);
> > >
> > > in PyDOLFIN
> > > bc = DirichletBC(function, mesh, subdomain, geometric)
> > >
> > > this gives
> > > NameError: name 'geometric' is not defined
> > >
> > > (same for pointwise)
> > >
> > > Jake
> >
> > Does it work if you instead write dolfin.geometric?
>
> Yes (see below).
>
> > In that case, it's a simple fix to include it in the
> > __init__.py file for DOLFIN (site-packages/dolfin/__init__.py).
>
> Adding the following to __init__.py got rid of the NameError.
>
> # BC specification
> from dolfin import topological, geometric, pointwise
>
> However a simple 2D DG elasticity script uncovers other problems.  Namely,
> trying to use the pointwise specification gives the following error:
>
>         TypeError: 'PySwigObject' object is unscriptable Swig director
> method error. Error detected when calling 'SubDomain.inside'

It seems that swig get problems when the function 

  user_sub_domain->inside(data.coordinates[i], false)

is called in

  DirichletBC::computeBCPointwise

Here data.coordinates is a double** array. The double* that is sent in to 
inside is wrapped by swig as a crude

  <Swig Object of type 'double *' at 0x86c44c0>

which is not subscriptable, and hence the error. 

I think this is a known problem as the dolfin devs have added a data type, 
simple_array, that wrap the double* and makes the array subscriptable. 

To Kent: Maybe something for your book chapter on known issues with 
crosslanguage programming? ;)

A patch that fix the problem is attached. :)

> (I have absolutely no idea what that even means, I don't know python/swig
> very well)
>
> Also, trying to use a LinearPDE gives a seg fault for the geometric
> specification.  (or some other nasty glibc error)

I haven't looked at this one...

Johan

# HG changeset patch
# User "Johan Hake <hake@xxxxxxxxx>"
# Date 1223926649 -7200
# Node ID d4b1ca4d9a980d8904de48988806b1ad1c5a1f69
# Parent  9e36520b1151b497e7c5ad21c9e1e0c3a78bbb7b
Fixed a bug when userdefined subdomain is used from python, together with pointwise applyied DirichletBC

diff -r 9e36520b1151 -r d4b1ca4d9a98 dolfin/fem/DirichletBC.cpp
--- a/dolfin/fem/DirichletBC.cpp	Mon Oct 13 00:16:19 2008 +0200
+++ b/dolfin/fem/DirichletBC.cpp	Mon Oct 13 21:37:29 2008 +0200
@@ -5,7 +5,7 @@
 // Modified by Martin Sandve Alnes, 2008
 //
 // First added:  2007-04-10
-// Last changed: 2008-10-01
+// Last changed: 2008-10-13
 
 #include <dolfin/common/constants.h>
 #include <dolfin/function/Function.h>
@@ -569,7 +569,8 @@
     for (uint i = 0; i < data.dof_map->local_dimension(); ++i)
     {
       // Check if the coordinates are part of the sub domain
-      if ( !user_sub_domain->inside(data.coordinates[i], false) )
+      simple_array<double> x(data.ufc_mesh.geometric_dimension, data.coordinates[i]);
+      if ( !user_sub_domain->inside(x, false) )
         continue;
       
       if(!interpolated)